基于支持向量机的香水识别电子鼻系统设计

针对电子鼻在化妆品领域的应用推广问题, 研制了一套用于香水识别的电子鼻系统。该系统主要包括数据采集, 信号调理, 特征提取和模式识别。硬件以微控制器dsPIC30F6014A为核心, 根据香水气味合理地选择了3个广谱型气敏传感器, 并加入温湿度传感器； 软件利用基于COM组件的VC和Matlab混合编程, 设计并实现了内嵌模式识别算法的上位机监控软件, 并发布成可执行文件。模式识别采用支持向量机（SVM: Support Vector Machine)方法, 相比PCA(Principal Component Analysis)、 神经网络等算法提高了小样本情况下的泛化性能。利用该系统对4种香水进行训练、 识别, 分类准确率达92%, 检测结果表明, 该电子鼻系统在香水识别方面具有较高的准确性和稳定性。

Design of Electronic Nose System for Based on Support Vector Perfume Recognition Machine

An electronic nose （e-nose） system has been designed in order to extend its application in the cosmetic field.The system is composed of four main parts including data acquisition, sensor signal conditioning, feature extraction and pattern recognition. The core of hardware used for data acquisition and transmission is dsPIC30F6014A. The sensor arrays are composed of three commercially available metal oxide semiconductor sensors by Figaro with an integrated temperature and humidity sensor. The PC control software of the e-nose system embedded with pattern recognition algorithm was developed by mixed programming between VC and Matlab based on COM component and made available as an executable file. Pattern recognition is based on SVM （Support Vector Machine）, which enhances the generalization performance of small sample compared to other methods such as PCA （Principal Component Analysis）, Neural Net and so on. The experimental results indicate that it is feasible to classify four kinds of perfume with up to 92% accuracy and identify the unknown sample using the e-nose system.